Lumian Gen AI Newsletter Issue #52

Welcome to the 52nd edition of the Lumian Weekly Gen AI Newsletter!

OpenAI just dropped their new “Strawberry” model, officially named o1, and it’s got everyone talking — or, well, reasoning. If you're keeping score at home, this is a pivot for OpenAI, a deliberate move from the typical pattern-recognition game that defines AI today into something that feels more like actual thinking. Or at least that’s the pitch.

Here’s the deal: o1 doesn’t throw an answer at you in one step, the way language models typically do. Instead, it takes the scenic route, solving problems step by step like it’s back in math class, showing its work. It’s slower, sure, and notably more expensive, but also supposedly way smarter when it comes to complex problems — the kind GPT-4o might fumble, like coding challenges or intricate math puzzles that make your head spin.

But here’s the thing — and this is classic AI PR — OpenAI is still hedging. The model’s not replacing GPT-4o; it’s complementing it. They’re quick to admit it’s slower, can’t browse the web, and isn’t multimodal (no image recognition here). So, if you’re thinking, “Okay, but can it explain why the stock market’s been bouncing like a pinball lately?” the answer is… probably not, unless your query is wrapped in some kind of elaborate math problem or a nested programming loop.

Still, OpenAI’s betting big on reasoning. They’ve even gone back to “1” in the model’s name to signal a reset in how they approach AI. And yes, the costs are steep — $15 per million input tokens and $60 per million output tokens, compared to GPT-4o’s far more budget-friendly rates. Essentially, if you’re running a startup and planning to feed this thing complex data all day, hope you’ve got a spare kidney to sell.

What’s especially interesting here is that OpenAI’s o1 seems designed to feel human — or at least give the illusion of being human. It throws in little phrases like “I’m thinking through…” and “Let me see…” like it’s your friend helping you solve a puzzle. But don’t get too excited — OpenAI is quick to remind you that it’s not really thinking; it’s just mimicking the process of thinking.

Which brings us back to the bigger question: what does this all mean? Well, it’s a shift. AI researchers have been chasing reasoning for a while now because if an AI can truly reason, it could open doors to breakthroughs in fields like medicine, engineering, and finance. You know, the stuff where small mistakes actually matter.

But until then, what we’ve got is an impressive — if expensive — incremental upgrade. It’s less about how many questions o1 can answer and more about how it can explain its answers. Which, if you ask me, makes it a bit like assembling IKEA furniture: slow, detailed, and by the time you get to the end, you’re not entirely sure why it took so long, but hey, it works.

Happy reading! 📚🤖🎵

In this week’s issue:

• News Flash: Klarna Fires Salesforce / Workday, Apple Intelligence, Replit Agents

• AI Frontier: AI Task Management tools you can use today

• Fundraising: The biggest deals in AI

• Nerd Out: Technical and Business Content for Everyone

🚀 AI in Practice

Cutting-Edge AI Task Management Tools You Can Use Today

• Hoop - AI task management for busy professionals

• Taskade - Efficiently manage your projects and teams